How to put the large set of the AMBRE model templates in github for flux calibration

Initially, we plan to store the 6k templates (2.4GB) in GitHub for flux calibration. Because this might be large, Mineo-san and Yamashita are discussing reducing the size.
One option is to upload a small subset of them. This enables us to run a test code for the blue-part in flux calibration but the output is inaccurate. The full data will be uploaded sometime in the future.
Another option is to compress the 6k templates. We are now discussing whether saving them in fixed-point numbers works well or not.
 
We have ~6k templates. In the current plan, a user generates a large template set (5.6k) with smaller parameter grids for stellar typing using the 6k templates and RBF interpolation in the f_star repository, on a user-side. Stellar typing for flux calibration refers to the 5.6k templates. 1 template (1 FITS file) has ~400kB. The size of the first template set of 6k is 6k*400kB = 2.4 GB in total.